Predictive Validity of the Postpartum Depression Predictors Inventory-Revised

Article Outline

• Summary
• Introduction
• Methods
  • Study participants
  • Instruments
    • PDPI-R
    • EPDS
  • Data collection
  • Data analysis
• Results
  • General characteristics of the participants
  • Distribution of the PDPI-R and EPDS scores
  • Discrimination
  • Calibration
  • Uniformity of fit
• Discussion
• Conclusion
• Conflict of interest
• Acknowledgments
• References
• Copyright

Summary 

Purpose

This study aimed to investigate the predictive validity of three versions of the Postpartum Depression Predictors Inventory-Revised (PDPI-R) in Korea.

Methods

A descriptive cross-sectional design with a self-administered questionnaire, including 43 items of the PDPI-R, using the Edinburgh Postpartum Depression Scale as the gold standard was used. Data were collected from 316 women within 6 weeks after childbirth in Busan, Korea, from August to November 2010.

Results

The postpartum depression and postpartum depressive symptom (PDS) rate was 22.5%. The area under the curve of the receiver operating characteristic curve was .882 for the prenatal version of the PDPI-R and .927 for the full version. The sensitivity and specificity were 87.3% and 85.1%, respectively, at a cutoff point of 9.5 for the full version, and 91.5% and 66.1%, respectively, at a cutoff point of 5.5 for the prenatal version. The Hosmer-Lemeshow goodness-of-fit statistics was 3.554 (p=.829) for the prenatal version and 8.305 (p=.404) for the full version; this showed a good degree of correspondence between the estimated and observed probabilities of PDS. By age, education, and socioeconomic groups, the discrimination and calibration were generally good for both the prenatal and full versions.

Conclusion

The PDPI-R showed good predictive validity among women in Korea. It is recommended that the prenatal version of the PDPI-R be used to predict PDS for pregnant women and the full version of the PDPI-R be used for women during the postpartum period.

Keywords: depression, predictive value of tests, postpartum period

Back to Article Outline

Introduction 

Postpartum depression or postpartum depressive symptom (PDS), a depression or depressive symptom that develops postnatally or starts before childbirth and continues after birth, contributes greatly to postnatal maternal suicide and is a serious problem in the field of health and medicine. PDS is one of the most universal health problems worldwide. According to systematic literature reviews, the prevalence of major depressive episodes is around 23% in Korean pregnant women (Ryu, Kim, & Lee, 2010), 6.5–12.9% until the first month after childbirth, 19.2% until the third month (Gavin et al., 2005). PDS may occur anytime during the first year after childbirth (Beck & Gable, 2001). However, according to a longitudinal study, it increases gradually until the sixth week after childbirth and then begins to decrease (Gjerdingen, Frober, Chaloner, & McGover, 1993).

This prevalence, however, is highly likely to be an underestimation. It is reported that around 50% of women who have experienced PDS do not recognize negative emotions arising from it nor do they report them to health practitioners (McGill, Burrows, Holland, Langer, & Sweet, 1995). Thus, essential prevention strategies are required to identify high-risk parturient women early by assessing the risk of PDS in pregnant women prenatally using a highly predictable tool (Austin & Lumley, 2003). Various scales have been developed in order to assess the risk of PDS in pregnant women. Representative tools developed relatively include the Vulnerable Personality Style Questionnaire (Boyce, Hickey, Gilchrist, & Talley, 2001), Postpartum Depression Predictors Inventory–Revised (Beck, 2001), Brisbane Postnatal Depression Index (Webster, Pritchard, Creedy, & East, 2003), Pregnancy Risk Questionnaire (Webster et al., 2003), and Contextual Assessment of Maternity Experience (Bernazzani et al., 2005).

All the PDS prediction tools mentioned above, excluding the Vulnerable Personality Style Questionnaire include both prenatal and postnatal factors as items related to PDS; thus, they may enable minimizing the condition via adequate management of prenatal factors during pregnancy. In particular, the Postpartum Depression Predictors Inventory-Revised (PDPI-R) includes all of the factors found to be significantly related to PDS via a systematic literature review. Thus, it covers relatively more diverse aspects of PDS compared to other tools which can minimize the omission of parturient women with a high risk of PDS. In addition, when the EPDS is used as the gold standard, it shows the highest sensitivity at a certain cutoff point. This study chose the PDPI-R as its PDS prediction tool, and evaluated its usability in Korea by testing its predictive validity.

On the other hand, in order for a PDS prediction tool to measure PDS accurately, it should satisfy three factors: discrimination, calibration, and uniformity of fit (Zimmerman et al., 1998). Discrimination is the ability to distinguish those with PDS from those without it, and is evaluated with the area under the curve (AUC) of the receiver operating characteristic (ROC) curve (Hanley & McNeil, 1983), as well as with validity indicators such as sensitivity, specificity, and predictability, and the correct classification rate. Calibration is a scale of agreement between PDS estimated by the prediction tool and the actual PDS, and is evaluated by analyzing the agreement between the number of predicted PDS women and the number of observed PDS women, standardization rate, and results of the Hosmer-Lemeshow goodness-of-fit (H-L fit) test (Hosmer & Lemeshow, 1989, pp. 82–134). Uniformity of fit means uniformity in the predictive validity of the PDS prediction tool among subgroups, and is evaluated by measuring discrimination and calibration according to the subjects’ general characteristics (Zimmerman et al.).

The aim of this study was to test the predictive validity of the PDS prediction tool PDPI-R, using the EPDS as the gold standard in parturient women 3–6 weeks after childbirth and who visited gynecology clinics or women’s hospitals in the Busan Metropolitan city. The specific objectives of this study were (a) to survey the participants’ general characteristics and their PDS levels, (b) to measure the discrimination of PDPI-R, (c) to measure the calibration of PDPI-R, and (d) to measure the uniformity of fit of PDPI-R according to the participants’ general characteristics.

Back to Article Outline

Methods 

Study participants 

Depression occurring 4 weeks after giving birth was defined as PDS by the Diagnostic and Statistical Manual of Mental Disorders, fourth edition (DSM-IV; American Psychiatric Association, 1994). Due to its higher sensitivity and specificity, the EPDS test was recommended to be administered within 6 weeks after birth (Cox, Holden, & Sagovsky, 1987). Therefore, the participants of this study were 316 women who had delivered not more than 6 weeks before and were recruited from women’s hospitals, or postpartum care facilities in Busan, Korea.

Instruments 

The PDPI-R, a screening tool for women at risk for developing PDS, developed from the findings of an updated meta-analysis of the risk factors of PDS (Beck, 2001, Beck, 2002, Beck et al., 2006). The PDPI-R consisted of 13 predictors of PDS: 10 prenatal factors and 3 postnatal factors. The prenatal factors were marital status, socioeconomic status, self-esteem, prenatal depression, prenatal anxiety, unwanted/unplanned pregnancy, history of previous depression, social support from husband/mate, family and friends, marital satisfaction, and life stresses (e.g., marital changes, occupational changes, crises). Postnatal factors included child care stress, infant temperament, and maternity blues. The prenatal version of the PDPI-R, which had a scale ranging from 0 to 32, was used during pregnancy. The full version of PDPI-R including both prenatal and postnatal factors, which had a scale ranging from 0 to 39, was used after delivery. The higher the score, the more risk factors a woman had for developing PDS (Beck et al., 2006).

Although the PDPI-R was designed to be administered via an interview conducted by a clinician (Beck & Gable, 2001), after its successful use in Australian women it could be used as a self-report questionnaire (Hanna, Jarman, Savage, & Layton, 2004). One of the strengths of the PDPI-R was that it could assess both the prenatal and postnatal risk for developing PDS, while most screening instruments for developing PDS were devised to be used only during pregnancy. The risk factors identified in the tool could also be used to develop intervention programs.

For this study, one of the researchers translated the instrument into Korean after getting permission from the developer (C. T. Beck), and a nursing researcher bilingual in Korean and English translated it into English. Coincidence between translation and back-translation was very high because the items of PDPI-R were short and simple. Next, we submitted the translated version of PDPI-R to a nursing professor and three physicians specializing in maternal care and familiar with the PDS lexical and content validation. Nursing professor and three physicians showed almost 100% of content validity index and asked us to make minor modifications such as typo correction or smooth expression without change of meaning. We also conducted a pilot test with five women who were within 6 weeks after delivery for the feasibility of the translated version of PDPI-R. The participants of pilot test asked to divide family support into the husband’s and wife’s sides of the family, in keeping with the Korean culture. We revised the instrument accordingly. Therefore, the prenatal version of the PDPI-R had 38 items with 10 risk factors from 0 to 3, and the final version of the PDPI-R had 45 items with 13 risk factors ranging from 0 to 45.

In this study, the level of PDS was assessed using the EPDS (Cox et al., 1987), which was one of the most widely used instruments for approximating PDS (Boyd, Le, & Somberg, 2005).

The EPDS consisted of 10 items for assessing a woman’s experience during the previous 7 days, and not just how she felt on the day of the survey. Each item was measured on a 4-point scale with a range of 0–3 (resulting range, 0–30); the higher the score, the higher the level of PDS. The EPDS had been validated for use either in pregnancy (Milgrom, Ericksen, Negri, & Gemmill, 2005) or postpartum. In Korea, the EPDS had been translated and validated with Korean women (Han et al., 2004, Kim et al., 2008). Han et al. (2004) showed a 100% sensitivity, a 78.4% specificity, and a 45.8% correct classification rate at a 9.5 cutoff point for both minor and major PDS, and a 100% sensitivity, a 90.5% specificity, and a 75.0% correct classification rate at a 12.5 cutoff point for major PDS only. Kim et al. (2005) determined the optimal cutoff point of EPDS for 236 pregnant women in Korea, resulting in a 79.7% sensitivity, an 87.1% specificity, and a .818 AUC of the ROC curve at 9.5 (9/10), the best cutoff point. According to the previous findings of Korean studies, we used 9.5 (9/10) as the cutoff point for the indication of probable PDS and divided the participants into a postpartum group (≥ 9.5) and a nonpostpartum group (<9.5). We used the Korean version of EPDS validated by Kim et al. (2005), and the internal consistency measured with Cronbach’s alpha was .851 in this study.

Data collection 

Data were collected from August to November 2010 after receiving an ethics approval from the Institutional Review Board, which is affiliated with the school to which the researchers belong. At first, we made a list of the hospitals specializing in women’s care, and postpartum care facilities located in the Busan area, and contacted the directors of the nursing departments of each of these institutions to get permission for the study. Finally, we selected 3 institutions with a consideration of the number of delivery and outpatient visitors within a month among the 10 institutions that agreed to join this study. One of these institutions had 200–250 deliveries per month, the other had 100–150 deliveries per month, and another facility gave postpartum care to an average of 40–80 women.

With permission from the directors of the nursing departments, one researcher visited the two hospitals to distribute and collect the questionnaires, and mailed the questionnaires, with return envelopes, to the director of the postpartum care facility. The director of the nursing department of the care facility was responsible for distributing and retrieving the questionnaires.

Data analysis 

Collected data were coded and analyzed using SPSS version 18.0 for Windows (SPSS Inc., Chicago, Il, USA). All statistical tests were performed a .05 level of significance for two-tailed tests.

The general characteristics of the participants were analyzed with means and standard deviations for continuous variables, and frequencies and percentages for categorical variables. The means and standard deviations of the PDPI-R and EPDS scores were obtained in total and by characteristics. The t test or analysis of variance test was used to see the differences in the PDPI-R and EPDS scores characteristics. The probabilities of PDS expected by the PDPI-R were compared with those observed by the EPDS (EPDS ≥ 9.5). The relative risks for PDS by general characteristics were presented with odds ratios (ORs) and 95% confidence intervals (CIs).

The discriminative power of the PDPI-R was measured with the AUC of the ROC which was a measure of the overall discriminatory power of the prognostic model in distinguishing those who had PDS from those who did not (Hanley & McNeil, 1983). Next, we calculated the sensitivity (true positive), specificity (true negative), positive predictive value (1-false positive), and negative predictive value (1-false negative) at certain cutoff points with the ROC curve.

The H-L fit statistic in the logistic regression model was used to evaluate the calibration of the PDPI-R (Hosmer & Lemeshow, 1989, pp. 82–134). Participants were divided into approximately 10 groups of roughly the same size based on the percentiles of the estimated probabilities. This goodness-of-fit statistic has a chi-square distribution and the discrepancies between the observed and expected number of observations in these groups were summarized by the Pearson chi-square statistic and p value (Hosmer & Lemeshow). Percentage agreement and kappa between the observed and expected number of PDSs were obtained. Standardized mortality ratios (SMR) were obtained by dividing the observed number by the expected number of PDSs for each category, and 95% CI were calculated.

For evaluating the uniformity of fit in different subgroups, we stratified patients by age group (< 30 yr vs. ≥ 30 yr), education level (high school and below vs. college and above), and perceived socioeconomic status (low vs. middle or high level). The discrimination and calibration of the PDPI-R were reanalyzed in each subgroup.

Back to Article Outline

Results 

General characteristics of the participants 

The average age of the participants was 31.5 years old, 77.2% graduated from college and above, and 85.1% reported that they were of middle or high level perceived socioeconomic status (Table 1).

Table 1. Distribution of General Characteristics, Postpartum Depression Predictors Inventory-Revised (PDPI-R), and Korean Version of the Edinburgh Postnatal Depression Scale (EPDS) Scores (N=316).

Characteristics		n (%)	PDPI-R (M±SD)		EPDS
			Prenatal version	Full version	M±SD	% ≥9.5	Relative risk (95% CI)
Total			6.3±5.0	8.3±5.4	6.4±4.6	22.5
Age (yr) (M±SD=31.5±3.6)	<30	95 (30.1)	7.3±6.1	9.3±6.9	7.2±5.6	32.6	1.80 (1.21–2.70)
	≥30	221 (69.9)	5.8±4.3	7.9±4.6	6.0±4.0	18.1	1
	t/χ2		2.076	1.824	1.796	8.05
	p		.040	.070	.075	.005
Education level	≤High school	72 (22.8)	8.0±6.0	10.4±6.6	8.1±5.3	33.3	1.73 (1.14–2.62)
	≥College f	244 (77.2)	5.8±4.5	7.7±4.9	5.8±4.3	19.3	1
	t/χ2		2.948	3.211	3.729	6.319
	p		.004	.002	<.001	.012
Perceived SES	Low level	47 (14.9)	9.7±6.0	12.2±6.6	9.7±5.2	46.8	1.54 (1.17–2.02)
	High or middle level	269 (85.1)	5.7±4.5	7.7±4.9	5.8±4.3	18.2	1
	t/χ2		−5.368	−4.484	−5.583	18.777
	p		<.001	<.001	<.001	<.001
Time after delivery (wk)	3	83 (26.3)	7.4±5.0	8.8±5.4	6.8±5.0	30.1	2.50 (1.10–5.67)
	4	73 (23.1)	6.7±4.7	9.2±5.4	7.5±4.6	30.1	2.50 (1.08–5.78)
	5	92 (29.1)	5.6±5.0	7.9±5.3	5.9±4.4	15.2	1.04 (0.43–2.51)
	6	68 (21.5)	5.3±5.0	7.3±5.6	5.3±4.2	14.7	1
	F/χ2		3.092	1.925	3.361	10.383
	p		.027	0.13	.019	.016
	Tukey test	3 wk>6wk (p=.047)			4 wk > 6 wk (p=.22)

Note. Tukey method was used for multiple comparisons in “Time after delivery”. SES=socioeconomic status.

Distribution of the PDPI-R and EPDS scores 

The distribution of the PDPI-R and EPDS scores and their relationship with the general characteristics were summarized in Table 1. The average PDPI-R score was 6.3 in the prenatal version and 8.3 in the full version. The average PDPI-R score of the prenatal version was related to age (p=.040), education level (p=.004), perceived socioeconomic status (p<.001), and weeks after delivery (p=.027), while the average PDPI-R score of the full version was related to education level (p=.002) and perceived socioeconomic status (p<.001).

The average EPDS score was 6.4 (21.3 based on 100 points), and was related to education level (p<.001), perceived socioeconomic status (p<.001), and weeks after delivery (p=.019). In total, 22.5% of the participants showed a 9.5 or higher score, and 10.8% received 12.5 or higher on the EPDS. PDS was significantly higher for participants under 30 years of age (OR=1.80, 95% CI=1.21–2.70), with a high school and below education level (OR=1.73, 95% CI=1.14–2.62), and with a lower perceived socioeconomic status (OR=1.54, 95% CI=1.17–2.02) than that of their counterparts. PDS was significantly higher at 3 or 4 weeks after delivery than at 5 or 6 weeks after delivery (Table 1).

Discrimination 

The discrimination findings of the prenatal version and full version of the PDPI-R were shown in Table 2 and Fig. 1. The AUC of the ROC curve for the prenatal version of the PDPI-R was .882 (95% CI=.838–.925), while the sensitivity, specificity, positive predictability, negative predictability, and correct classification rate were 91.5%, 66.1%, 43.9%, 96.4%, and 71.8%, respectively, at a 5.5 (5/6) cutoff point.

Table 2. AUC of the ROC Curve, Sensitivity, and Specificity at Certain Cutoff Points.

Note. AUC=area under the curve; ROC=receiver operating characteristic; CI=confidence interval; PDPI-R=Postpartum Depression Predictors Inventory-Revised.

• 
  • View Large Image
  • Download to PowerPoint

• Fig. 1 

  Receiver operating characteristic curve: Postpartum Depression Predictors Inventory-Revised (PDPI-R) on the Edinburgh Postnatal Depression Scale (EPDS). Note. AUC=area under the curve.

The AUC of the ROC curve for the full version of the PDPI-R was .927 (95% CI=.893–.961), while the sensitivity, specificity, positive predictability, negative predictability, and correct classification rate were 90.1%, 82.0%, 59.2%, 96.6%, and 83.8%, respectively, at a 9.5 (9/10) cutoff point.

Calibration 

The calibration findings of the prenatal and full versions of the PDPI-R were shown in Table 3.

Table 3. Hosmer-Lemeshow Goodness-of-fit Statistics and Agreement Between Observed and Expected Postpartum Depression or Postpartum Depressive Symptom (PDS).

Note. O/E=observed/expected; PDPI-R=Postpartum Depression Predictors Inventory-Revised.

The value (chi-square) of the HL fit statistic was 3.554 (p=.829) of the prenatal version of the PDPI-R and 8.305 (p=.404) for the full version of the PDPI-R. Because the null hypothesis of the HL fit test was that the model fits well, these results mean that the expected number of PDSs from this PDPI-R equation and the observed number of PDSs from the EPDS were similar for each category.

The percent agreement between the expected and observed number of PDSs was 85.1% for the prenatal version of the PDPI-R, and 88.0% for the full version of the PDPI-R.

The SMR, which was obtained by dividing the observed number by the expected number of PDSs for each category of the PDPI-R score, was .96 (95% CI=0.79–1.12) for the version of the PDPI-R and .99 (95% CI=0.72–1.25) for the full version of the PDPI-R.

Uniformity of fit 

In participants stratified according to age group, education level, and socioeconomic status, discrimination was generally good in all subgroups with a range from .863 to .947 in the AUCs of the ROC curve, while the chi-square of the H-L fit statistic showed a good fit for all subgroups. Most percentage agreement was over 80%, except among those participants less than 30 years old, for the prenatal version of the PDPI-R. Considering the number of observed and expected PDSs, the prenatal version of the PDPI-R underestimated PDS in participants over 30 years of age, and the full version of the PDPI-R underestimated PDS in participants less than 30 years of age (Table 4).

Table 4. AUC of the ROC Curve, Hosmer-Lemeshow Goodness-of-fit Statistics, and Agreement Between Observed and Expected Postpartum Depression or Postpartum Depressive Symptom by General Characteristics.

Characteristics		Hosmer-Lemeshow goodness- of-fit statistics	Observed and expected PDS
		AUC of ROC curve (95% CI)	χ2	p	Percent agreement (kappa)	O/E (95% CI)
PDPI-R prenatal version
Age (yr)	<30	.869 (.797–.941)	4.458	.814	78.9% (.62)	0.96 (0.77–1.15)
	≥30	.890 (.833–.947)	5.809	.562	87.3% (.63)	1.04 (0.62–1.46)
Education	≤High school	.892 (.821–.963)	5.817	.561	80.6% (.71)	0.89 (0.60–1.18)
level	≥College	.872 (.579–.760)	3.208	.921	86.5% (.59)	0.99 (0.73–1.24)
Perceived	Low level	.901 (.818–.980)	2.663	.914	80.9% (.70)	0.86 (0.58–1.13)
SES	High or middle level	.863 (.807–.919)	4.911	.767	85.9% (.58)	0.95 (0.74–1.16)
PDPI-R full version
Age (years)	<30	.926 (.873–.979)	3.595	.825	86.3% (.62)	1.03 (0.68–1.39)
	≥30	.928 (.885–.972)	9.487	.303	89.6% (.63)	0.95 (0.60–1.30)
≤High school
Education	≤High school	.947 (.897–.996)	6.976	.539	87.5% (.71)	0.82 (0.50–1.13)
level	≥College	.916 (.871–.960)	3.911	.865	88.1% (.59)	0.98 (0.73–1.22)
Perceived SES	Low level	.922 (.844–.999)	3.552	.830	85.1% (.70)	0.93 (0.54–1.32)
	High or middle level	.918 (.875–.961)	5.528	.596	88.5% (.58)	0.92 (0.67–1.16)

Note. AUC=Area under the curve; ROC=receiver operating characteristic; CI=confidence interval; O/E=observed/expected; PDS=postpartum depression or postpartum depressive symptom; SES=socioeconomic status.

Back to Article Outline

Discussion 

This study tested the predictive validity of the PDS prediction tool PDPI-R, using the EPDS as the gold standard, with 316 parturient women 3–6 weeks after childbirth.

According to the results of measuring with the EPDS, 22.5% of the participants received a score of 9.5 or higher, while 10.8% received 12.5 or higher. In their systematic literature review of theses published between 1980 and 2004, Gavin et al. (2005) reported that the prevalence of major depressive episodes until 3 months after childbirth was 19.2%; a study in Jordan reported that 22% of women at 6–8 weeks postpartum received a score of 13 or higher (Mohammad, Gamble, & Creedy, 2010); and a study in the United States reported that 13% of women at 4–8 weeks postpartum received a score of 10 or higher (Horowitz, Murphy, Gregory, & Wojcik, 2011). While all of these studies measured PDS using the EPDS, they are not comparable with one another because they measured EPDS at different period after childbirth and used different cutoff points. Cox et al. (1987), who developed the tool, suggested 12.5 as a cutoff point, but studies in Western Europe used various cutoff points ranging from 9.5 to 12.5. In contrast, a study by Ryu et al. (2010) showed that 23% of women during the postpartum period were regarded as having PDS, similar to our finding at the same cutoff point. Considering that Korean studies such as those by Han et al. (2004) and Kim et al. (2008) suggested 9/10 as the cutoff point while Japanese (Yoshida et al., 1997) and Chinese ones (Lee et al., 1998) suggested 10.5 and 9.5, respectively, it is adequate to use 9/10 as the standard cutoff in future research in Korea.

In order to test predictive validity in this study, we evaluated the discrimination, calibration, and uniformity of fit calibration according to the general characteristics. Discrimination is the ability to distinguish those with PDS from those without it, and is generally measured with the AUC of the ROC curve (Hanley & McNeil, 1983). The AUC is evaluated to be “noninformative (AUC=.5), less accurate (.5<AUC≤.7), moderately accurate (.7<AUC≤.9), highly accurate (.9<AUC<1.0), or perfect (AUC=1)” (Greiner, Pfeiffer, & Smith, 2000). In our study, the full and prenatal versions of the PDPI-R had ROC AUCs of .927 and .882, respectively, showing relatively substantial discrimination.

The cutoff point of the PDPI-R was estimated through ROC analysis. Because it is very important for a PDS prediction tool to identify parturient women with a high risk of PDS early, it is desirable to raise sensitivity as high as possible while maintaining a level of specificity so that all pregnant women suspected to have the risk of PDS can be detected. Sensitivity is the proportion of subjects who had PDS in the EPDS that was expected correctly by model of the PDPI-R score; specificity is the proportion of subjects who did not have PDS in the EPDS that was expected correctly by model of the PDPI-R score. In terms of this criterion, the optimal cutoff point of the PDPI-R based on the results of this study were 5.5 for the prenatal version and 9.5 for the full version. In a study by Beck et al. (2006), who developed the inventory, the prenatal version of the PDPI-R explained 67% of the variance of PDS and its sensitivity, specificity, and ROC AUC was 76%, 54%, and .673 at a cutoff point of 10.5 using the EPDS as the gold standard PDS, respectively (Beck et al., 2006). We cannot explain the exact reason of the big difference in cutting score between the two studies. However, the study of Beck et al. (2006) showed “less accuracy” in AUC and lower level of specificity than that of our study, so we recommend 5.5 as a cutoff point of prenatal version of PDPI-R and further studies to confirm the cutoff point of PDPI-R. In the prospective study by Oppo et al. (2009), in which interviews using the DSM-IV were used as the gold standard, the prenatal version showed a sensitivity of 76% and a specificity of 71% at a cutoff of 3.5, while the full version showed a sensitivity and specificity of 82% and 75%, respectively, at a cutoff point of 5.5, demonstrating large differences in its cutoff points from the results of Beck et al. (2006). The possible reason for the differences in studies is the tool used for the measurement of PDS. Our study and the Beck et al. (2006) study determined PDS by measuring the EPDS, but Oppo et al. (2009) determined whether a subject belonged to the category of DSM-IV by consulting a psychiatrist or trained psychological counselor if the score of the EPDS was 13 or higher. Beck et al. (2006) also determined whether a subject belonged to the category of DSM-IV by consulting a psychiatric nurse for subjects with a high EPDS score. In other words, these previous studies used different criteria for the final diagnosis of PDS.

Calibration is a measure of agreement between the number of PDS women predicted by a PDS prediction tool and the number of actual PDS women. In this study, all three versions of the PDPI-R showed high calibration. In the results of the H-L fit test using a logistic regression model, all three versions were fit for the model. That is, PDS predicted by the PDPI-R showed high agreement with actual PDS evaluated with the EPDS. In addition, the kappa value obtained by correcting accidental agreement in the calibration between predicted and observed PDS was a substantial .63 for all versions. In general, a kappa value of .21 or higher is considered fair, .41–.60 moderate, and .61–.80 substantial (Viera & Garrett, 2005).

In order to test the uniformity of the predictive validity of the PDS prediction tool according to subgroups, we assessed the discrimination calibration according to the participants’ general characteristics, such as age, education level, and socioeconomic level. The full version of the PDPI-R showed a substantial uniformity of discrimination among the subgroups according to the three general characteristics. The uniformity of calibration also showed a satisfactory level for all of the characteristics. The prenatal version of the PDPI-R showed a somewhat lower AUC of the ROC curve than the full version did, but it was still over .85, showing that its discrimination was acceptable. Its calibration was also satisfactory.

This study is meaningful in that it suggested the optimal cutoff point in using the PDPI-R and the clinical utility of the tool, but it has a number of limitations to be considered while interpreting the results. First, because the study used self-reported responses, the data may not be accurate. Second, because it surveyed both prenatal and postnatal factors only postpartum, there may be a recall bias.

Back to Article Outline

Conclusion 

The PDPI-R showed good predictive validity among women in Korea. Therefore, we recommend that the prenatal version of the PDPI-R be used to predict PDS for pregnant women and the full version of the PDPI-R be used for women during the postpartum period. The recommended cutoff points were 5.5 (5/6) and 9.5 (9/10) for prenatal version of the PDPI-R and full version of the PDPI-R, respectively. We also recommend further study to verify these cutoff points of two instruments in a representative sample in Korea because our study was performed in the limited sample in one area of Korea.

Back to Article Outline

Conflict of interest 

The authors declare no conflict of interest.

Back to Article Outline

Acknowledgments 

We would like to thank the directors of the three study institutions as well as the participants for their time, and Dr. C.T. Beck for giving his permission to use the PDPI-R questionnaire for this study.

Back to Article Outline

References 

1. American Psychiatric Association . Diagnostic and statistical manual of mental disorders. 4th ed.. Washington, DC: Author; 1994;
   • View In Article
2. Austin MP, Lumley J. Antenatal screening for postnatal depression: a systematic review. Acta Psychatrica Scandinavica. 2003;107:10–17
   • View In Article
3. Beck CT. Predictors of postpartum depression: an update. Nursing Research. 2001;50:275–285
   • View In Article
   • MEDLINE
   • CrossRef
4. Beck CT. Revision of the postpartum depression predictors inventory. Journal of Obstetric, Gynecologic, and Neonatal Nursing. 2002;31:394–402
   • View In Article
5. Beck CT, Gable RK. Comparative analysis of the performance of the postpartum depression screening scale with two other depression instruments. Nursing Research. 2001;50:242–250
   • View In Article
   • MEDLINE
   • CrossRef
6. Beck CT, Records K, Rice M. Further development of the postpartum depression predictors inventory-revised. Journal of Obstetric, Gynecologic, and Neonatal Nursing. 2006;35:735–745
   • View In Article
7. Bernazzani O, Marks MN, Bifulco A, Siddle K, Asten P, Conroy S. Assessing psychosocial risk in pregnant/postpartum women using the Contextual Assessment of Maternity Experience (CAME). Social Psychiatry and Psychiatric Epidemiology. 2005;40:497–508
   • View In Article
   • MEDLINE
   • CrossRef
8. Boyce P, Hickey A, Gilchrist J, Talley NJ. The development of a brief personality scale to measure vulnerability to postnatal depression. Archives of Women’s Mental Health. 2001;3:147–153
   • View In Article
   • CrossRef
9. Boyd RC, Le HN, Somberg R. Review of screening instruments for postpartum depression. Archives of Women’s Mental Health. 2005;8:141–153
   • View In Article
   • MEDLINE
   • CrossRef
10. Cox JL, Holden JM, Sagovsky R. Detection of postnatal depression: development of the 10-item Edinburgh postnatal depression scale. British Journal of Psychiatry. 1987;150:782–786
    • View In Article
11. Gavin NI, Gaynes BN, Lohr KN, Meltzer-Brody S, Gartlehner G, Swinson T. Perinatal depression: a systematic review of prevalence and incidence. Obstetrics and Gynecology. 2005;106:1071–1083
    • View In Article
    • MEDLINE
    • CrossRef
12. Gjerdingen DK, Frober DG, Chaloner KM, McGover PM. Changes in women’s physical health during the first postpartum year. Archives of Family Medicine. 1993;2:277–283
    • View In Article
    • MEDLINE
13. Greiner M, Pfeiffer D, Smith RD. Principles and practical application of the receiver-operating characteristic analysis for diagnostic tests. Preventive Veterinary Medicine. 2000;45:23–41
    • View In Article
    • MEDLINE
    • CrossRef
14. Han KW, Kim MG, Park JM. The Edinburgh postnatal depression scale, Korean version: reliability and validity. Journal of the Korean Society of Biological Therapies in Psychiatry. 2004;10:201–207
    • View In Article
15. Hanley J, McNeil B. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology. 1983;148:839–843
    • View In Article
    • MEDLINE
16. Hanna B, Jarman H, Savage S, Layton K. The early detection of postpartum depression: midwives and nurses trial a checklist. Journal of Obstetric, Gynecologic, & Neonatal Nursing. 2004;33:191–197
    • View In Article
17. Horowitz JA, Murphy CA, Gregory KE, Wojcik JA. Community-based screening initiative to identify mothers at risk for postpartum depression. Journal of Obstetric, Gynecologic, and Neonatal Nursing. 2011;40:52–61
    • View In Article
18. Hosmer D, Lemeshow S. Model building strategies and methods for logistic regression. In:  Hosmer D,  Lemeshow S editor. Applied logistic regression. New York: John Wiley & Sons; 1989;
    • View In Article
19. Kim YK, Hur JW, Kim KH, Oh KS, Shin YC. Clinical application of Korean version of Edinburgh Postnatal Depression Scale. Journal of the Korean Neuropsychiatric Association. 2008;47:36–44
    • View In Article
20. Kim YK, Won SD, Lim HJ, Choi SH, Lee SM, Shin YC, et al. Validation of the Korean version of Edinburgh Postnatal Depression Scale (K-EPDS). Journal of Korean Society for Depression and Bipolar Disorders. 2005;3:42–49
    • View In Article
21. Lee DT, Yip SK, Chiu HF, Leung TY, Chan KP, Chau IO, et al. Detecting postnatal depression in Chinese women: Validation of the Chinese version of Edinburgh Postnatal Depression Scale. British Journal of Psychiatry. 1998;172:433–437
    • View In Article
22. McGill H, Burrows V, Holland L, Langer H, Sweet M. Postnatal depression: a Christchurch study. New Zealand Medical Journal. 1995;108:162–165
    • View In Article
23. Milgrom J, Ericksen J, Negri L, Gemmill AW. Screening for postnatal depression in routine primary care: Properties of the Edinburgh postnatal depression scale in an Australian sample. The Australian and New Zealand Journal of Psychiatry. 2005;39:833–839
    • View In Article
    • MEDLINE
    • CrossRef
24. Mohammad KI, Gamble J, Creedy DK. Prevalence and factors associated with the development of antenatal and postnatal depression among Jordanian women. Midwifery. 2010;27:238–245
    • View In Article
25. Oppo A, Mauri M, Ramacciotti D, Camilleri V, Banti S, Borri C, et al. Risk factors for postpartum depression: The role of the Postpartum Depression Predictors Inventory-Revised (PDPI-R). Results from the perinatal depression-research & screening unit (PNDReScU) study. Archives of Women’s Mental Health. 2009;12:239–249
    • View In Article
    • CrossRef
26. Ryu AL, Kim TH, Lee HH. Evaluation of risk factors for the postpartum depression with Edinburgh Postnatal Depression Scale (EPDS) Score. Korean Journal of Perinatology. 2010;21:74–80
    • View In Article
27. Viera AJ, Garrett JM. Understanding interobserver agreement: The kappa statistic. Family Medicine. 2005;37:360–363
    • View In Article
    • MEDLINE
28. Webster J, Pritchard MA, Creedy D, East CA. Simplified predictive index for the detection of women at risk for postnatal depression. Birth. 2003;30:101–108
    • View In Article
    • MEDLINE
    • CrossRef
29. Yoshida K, Marks MN, Kibe N, Kumar R, Nakano H, Tashiro N. Postnatal depression in Japanese women who have given birth in England. Journal of Affective Disorders. 1997;43:69–77
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (119 KB)
    • CrossRef
30. Zimmerman JE, Wagner DP, Draper EA, Wright L, Alzola C, Knaus WA. Evaluation of acute physiology and chronic health evaluation III predictions of hospital mortality in an independent database. Critical Care Medicine. 1998;26:1317–1326
    • View In Article
    • MEDLINE
    • CrossRef